研究人员正在探索在不增加模型规模或计算成本的情况下增强小型语言模型（SLM）推理能力的方法。一种方法侧重于推理前的提示消歧，识别并解决用户提示中的语义风险，以提高大型语言模型对关键标记的注意力，仅花费0.02美元即可带来2.5个点的性能提升。另一种策略是双轨CoT（Dual-Track CoT），旨在通过采用预算感知分步指导和控制冗余步骤，使小型语言模型能够在严格的标记和计算预算内可靠地执行多步推理。 AI
影响 新技术可能使资源受限环境中的小型语言模型实现更高效、更具成本效益的推理。
排序理由 该集群包含两篇arXiv论文，详细介绍了改进小型语言模型推理能力的新研究。
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2604.25039v1 Announce Type: new Abstract: Large Language Models (LLMs) solve many reasoning tasks via chain-of-thought (CoT) prompting, but smaller models (about 7 to 8B parameters) still struggle with multi-step reasoning under tight compute and token budgets. Existing tes…
arXiv:2604.23263v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly utilized in various complex reasoning tasks due to their excellent instruction following capability. However, the model's performance is highly dependent on the open-ended characteristic…
Large Language Models (LLMs) solve many reasoning tasks via chain-of-thought (CoT) prompting, but smaller models (about 7 to 8B parameters) still struggle with multi-step reasoning under tight compute and token budgets. Existing test time reasoning methods such as self consistenc…